AI puts on its thinking cap for materials engineering
AI is taking the world by storm, popping up in every nook and cranny of human endeavor through its ability to crunch huge amounts of data and come to speedy conclusions. Materials science and engineering is no exception, as many of its next-leap advances depend on sifting through large datasets to understand material structure, properties and applications. Arun Mannodi Kanakkithodi, assistant professor in the School of Materials Engineering, shares his thoughts on the subject in a Q&A with Purdue Engineering Review.
What is AI?
Artificial intelligence (AI) is a branch of computer science that aims to replicate or simulate human intelligence in machines. The origins of AI lie in Alan Turing’s “thinking machines,” which gave way to the “Turing test” for determining whether somebody (or something) is a computer or a human. Machine learning (ML) and data science are components, or sub-fields, of AI. The former deals with the study of computer algorithms that improve automatically with experience and by the use of data, while the latter concerns extracting knowledge from structured or unstructured data using scientific methods, processes, and algorithms.
How is AI used overall?
AI and ML are now indispensable components of all facets of society and the global economy, affecting everything from banking to healthcare to food to the fundamental way modern offices and businesses operate. Examples of functions in which AI is playing a crucial role — with or without people realizing — include determining housing prices based on region and population demographics, making complicated weather forecasts based on existing conditions and historical data, screening resumes for a job opening, issuing credit cards, and discovering new pharmaceutical drugs. Not every such application is ethical or necessary, and businesses still depend on people training, executing and deploying AI/ML models to make the right calls in their specific industries. At the end of the day, AI (if used correctly) generally helps save time by providing a way to navigate highly complex problems involving tons of data and high-dimensional spaces that the human mind cannot easily visualize.
How do AI and materials science and engineering mix?
AI has been a vital part of research and education in materials science and engineering (MSE) for almost two decades. However, using data science or correlations between inputs and outputs to help understand materials behavior has been around in some form or another for centuries. In one 20th-century example, the famous Hall-Petch equation relating grain size to the strength of a material was discovered in the 1950s by analyzing a collected dataset and uncovering a simple relationship. With the advent of massive computing power, and methods ranging from Gaussian processes to neural networks to generative models, AI/ML now can be performed on enormous datasets amassed over decades. This AI/ML can speed the discovery of new materials and capabilities in the laboratory much faster than before, when brute-force Edisonian approaches — repeated trial and error — were the norm. The emerging field of materials informatics studies the structure, behavior and interactions of natural and engineered computational systems to improve the understanding, use, selection, development and discovery of materials.
Can you cite use cases in materials engineering?
AI is performing autonomous robotic experiments in the lab much faster than can be done manually; an “active learning” scheme then determines the next experiment to conduct in order to reach the best solutions in the fewest experiments. AI is generating new material structures with desired properties for a variety of applications, such as batteries and electronics. It is using large language models to navigate through historical text-based datasets in efforts to accelerate discovery of new materials. And AI is training models to accurately predict atomic forces and energies (referred to as ML-based interatomic potentials) to hasten the simulation of materials dynamics.
What about your research?
I am a computational materials scientist applying quantum mechanics-based density functional theory (DFT) simulations and AI/ML to drive materials discovery. My research group has developed some of the largest computational datasets of materials with applications in solar cells, electronics, and quantum computing. These datasets have helped train ML models to predict materials properties on demand, accelerating real-world materials discovery by several orders of magnitude. The breakthroughs I am hoping for will make it even easier to learn and perform high-throughput DFT computations using shared computing resources. Quantum computing may provide the ultimate solution in this regard, performing first-principles simulations in a fraction of the time required with classical computing. All data and associated models are available open-source via the Materials Data Facility and nanoHUB, a nanotechnology repository housed at Purdue. Our research has major consequences for semiconductor R&D, ties in well with the U.S. government’s 2022 CHIPS and Science Act as well as with the long-standing Materials Genome Initiative, and contributes to achieving the Semiconductors@Purdue goal of educating the next generation of workforce leaders in semiconductors and microelectronics.
What challenges remain?
The primary challenge is making high-quality and high-accuracy materials data readily available. Funding agencies and professional societies are emphasizing the need to make materials data more “FAIR” — findable, accessible, interoperable and reusable. However, many researchers are reluctant to share their data publicly, especially when it contains failed experiments or computations (which provide huge opportunities for learning); this leads to unnecessary repeated efforts.
What’s next for AI in materials engineering?
AI/ML in materials science is going to be of paramount importance going forward as we deal with the ever-increasing quantities of data and computing power, as well as the needs of a world trying to tackle climate change, energy crises, and a host of social and economic problems. AI in materials engineering will be transformative for renewable energy, hydrogen generation and storage, CO2 removal, and other important approaches to deal with climate change. AI will help reduce duplicated efforts of materials researchers, and will break barriers between domain experts and data scientists. My vision is that all MSE undergraduate students take courses on elementary coding/programming, data science and analytics, statistics, machine learning, and materials informatics, early on in their curriculum. No matter what the students go on to do, this knowledge will prepare them well for the future.
Arun Mannodi Kanakkithodi, PhD
Assistant Professor
Principal Investigator, Mannodi Research Group: Data-Driven Materials Design
School of Materials Engineering
College of Engineering
Purdue University
